n comms 8000

ARTICLE

Received 9 Sep 2014 | Accepted 23 Mar 2015 | Published 12 May 2015

Widespread seasonal gene expression revealsannual differences in human immunity andphysiologyXaquin Castro Dopico1,*, Marina Evangelou1,*, Ricardo C. Ferreira1, Hui Guo1, Marcin L. Pekalski1,

Deborah J. Smyth1, Nicholas Cooper1, Oliver S. Burren1, Anthony J. Fulford2, Branwen J. Hennig2,

Andrew M. Prentice2, Anette-G. Ziegler3, Ezio Bonifacio4, Chris Wallace1,5,** & John A. Todd1,**

Seasonal variations are rarely considered a contributing component to human tissue function

or health, although many diseases and physiological process display annual periodicities.

Here we nd more than 4,000 protein-coding mRNAs in white blood cells and adipose tissue

to have seasonal expression proles, with inverted patterns observed between Europe and

Oceania. We also nd the cellular composition of blood to vary by season, and these changes,

which differ between the United Kingdom and The Gambia, could explain the gene expression

periodicity. With regards to tissue function, the immune system has a profound pro-

inammatory transcriptomic prole during European winter, with increased levels of soluble

IL-6 receptor and C-reactive protein, risk biomarkers for cardiovascular, psychiatric and

autoimmune diseases that have peak incidences in winter. Circannual rhythms thus require

further exploration as contributors to various aspects of human physiology and disease.

DOI: 10.1038/ncomms8000 OPEN

1 JDRF/Wellcome Trust Diabetes and Inammation Laboratory, Department of Medical Genetics, NIHR Cambridge Biomedical Research Centre, CambridgeInstitute for Medical Research, University of Cambridge, Wellcome Trust/MRC Building, Cambridge Biomedical Campus, Cambridge CB2 0XY, UK. 2MRCInternational Nutrition Group at MRC Unit The Gambia & London School of Hygiene & Tropical Medicine, Keppel Street, London WC1E 7HT, UK. 3 Institute ofDiabetes Research, Helmholtz Zentrum Munchen, Neuherberg, Forschergruppe Diabetes, Klinikum rechts der Isar, Technische Universitat Munchen,Ingolstaedter Landstr. 1, D 85764 Neuherberg, Germany. 4 CRTDDFG Research Center for Regenerative Therapies Dresden, Paul Langerhans InstituteDresden, Medical Faculty, Technische Universitat Dresden, Fetscherstrasse, 01307 Dresden, Germany. 5MRC Biostatistics Unit, Cambridge Institute of PublicHealth, Forvie Site, Robinson Way, Cambridge Biomedical Campus, Cambridge CB2 0SR, UK. * These authors contributed equally to this work. ** Theseauthors jointly supervised this work. Correspondence and requests for materials should be addressed to X.C.D. (email: [email protected]) or to J.A.T.(email: [email protected]).

NATURE COMMUNICATIONS | 6:7000 | DOI: 10.1038/ncomms8000 |www.nature.com/naturecommunications 1

& 2015 Macmillan Publishers Limited. All rights reserved.

Periodic seasonal changes have inuenced all life forms, asexemplied by seasonal physiology and behaviours acrossplant and animal species13. For example, reptile graft

rejection4 and level of gonadal hormones in squirrel monkeys5

display seasonal variation. In humans, many complex polygenicdiseases, including cardiovascular6,7, autoimmune8,9 andpsychiatric illnesses1012, have established seasonal patterns ofincidence and disease activity. Infectious disease seasonality iswell established in humans13, and it has been proposed thatan inborn physiological rhythm underlies the seasonality ofdiagnoses of infectious diseases and their pathologies14, but directevidence of such a system is lacking.Various biological processes show seasonal variation in

humans, including ones with important immunological roles,such as vitamin D metabolism15. The loss of skin pigmentation ashumans migrated out of Africa to more temperate and colderzones to increase sunlight-driven vitamin D production is a majorexample of the evolutionary adaption of humans to differentenvironments. Yet, how seasons might more broadly impact theunderlying molecular details of human physiology is unknown.Along these lines, we hypothesized that the anti-inammatorycircadian transcription factor, ARNTL (BMAL1)16,17, woulddisplay seasonal gene expression differences as daylight entrainscircadian rhythms in mammals1821. Tissue-specic molecularclocks control a diverse range of cellular processes22,23,inuencing the immune response2428.From ethnically and geographically diverse populations we

analysed mRNA expression levels in peripheral blood mono-nuclear cells and adipose tissue biopsies, full blood count data,and the circulating levels of inammatory protein biomarkers.

ResultsSeasonal ARNTL expression in the immune system. We rstanalysed ARNTL expression in peripheral blood mononuclear cells(PBMCs) from children (454 samples from 109 individuals) enroledinto the BABYDIET cohort from Germany29 (SupplementaryTable 1). ARNTL mRNA showed seasonal variation in expression(ANOVA, w22, P 1.04 10 23), peaking in the summer monthsof June, July and August (Fig. 1a). The difference between thewinter low and summer high in ARNTL expression was 1.5097-fold.Vitamin D receptor (VDR) expression was also higher in thesummer months (Fig. 1a). The housekeeping genes, B2M andGAPDH, often used as standards in gene expression analyses, didnot show seasonal variation (Fig. 1a). ARNTL showed the sameseasonal expression prole independently of whether blood wasdrawn during morning or afternoon clinic visits (Fig. 1b),suggesting that diurnal oscillations are not responsible for theseasonal differences in ARNTL expression.We then sought evidence for seasonality in known components

of the circadian clock. Seasonal variation was found in 9 of the 16clock genes tested: ARNTL, CLOCK, CRY1, CSNK1D, CSNK1E,NR1D2, RORA, TIMELESS30 and NFIL3 (which controls diurnalTh17 cell development in mice31) (Fig. 1c). Seven genes (CRY2,PER3, RORB, NPAS2, PER1, PER2 and NR1D1) did not showevidence for seasonal effects (Supplementary Table 3). Novelcomponents of the human circadian clock, as well as clock-targeted genes and pathways, are likely to be present among thegenes whose expression correlated with ARNTL (SupplementaryTable 2). Interestingly, the glucocorticoid receptor (NR3C1) had astrong positive correlation with ARNTL (Spearman r 0.819),with lowest expression in the winter (ANOVA, w22,P 5.05 10 19) (Fig. 1d). Glucocorticoids have anti-inammatory properties32 and SCN-controlled hormones arethought to be essential molecules for maintaining thesynchronicity of peripheral biological clocks33. In contrast toNR3C1, receptors for the prostaglandins (PTGDR, PTGIR and

PTGER4), leukotrienes (CYSLTR1) and oxoeicosanoids (OXER1)were more highly expressed in the winter in Germany. Receptorsfor adiponectin (ADIPOR1), estradiol (ESR2) and antidiuretichormone (CUL5) were more highly expressed in the summer(Fig. 1d). Other hormone receptors did not show any seasonalvariation in this data set.

Widespread seasonal gene expression in the immune system.Strikingly, we foundB23% of the genome (5,136 unique genes outof 22,822 genes tested) to show signicant seasonal differences inexpression in the BABYDIET data set (Fig. 2a and SupplementaryTable 3). Among the seasonal genes, two distinct anti-phasic pat-terns of gene expression were evident: 2,311 genes (2,922 uniqueprobes) had increased expression in the summer (dened as June,July and August, mean fold change 1.2572) while 2,826 genes(3,436 unique probes) were upregulated in the winter (dened asDecember, January, February, mean fold change 1.3150)(Fig. 2c, Supplementary Fig. 1), demonstrating that differenttranscriptional landscapes are present in the peripheral immunesystem during different seasons.The daily variables of mean ambient temperature and mean

sunlight hours both served as linear predictors of seasonality(Supplementary Fig. 2), suggestive of human environmentaladaptation.We replicated the observation in two independent data sets.

First, in a collection of PBMCs isolated from autoimmune type 1diabetes (T1D) patients (236 samples) from the United Kingdom,1,697 genes were found to exhibit seasonal expression (Fig. 3a andSupplementary Table 4). The majority of seasonally associatedtranscripts could again be identied as having summer or winterexpression proles, with seasonal patterns matching thoseidentied in the BABYDIET data set (Fig. 3b). This data setdemonstrated seasonal gene expression in adults, adding to theobservations made in samples from children (SupplementaryFig. 3 and Fig. 2).Secondly, we analysed gene expression data from a collection of

PBMCs from adult (1883 years old, mean age 45) asthmaticindividuals from diverse ethnic groups across Australia, UnitedKingdom/Ireland, United States and Iceland34. We separated theentire cohort into distinct geographical locations and observedseasonal gene expression in each (Fig. 3c and SupplementaryTables 58). Seasonal genes identied in the BABYDIET cohortmaintained their seasonal tropisms in the asthmatic patients(Fig. 3d). Most interestingly, in the Australian data set, thepreviously dened summer genes (increased in expression duringthe Northern hemisphere summer) were more highly expressedduring the Southern hemisphere summer, spanning December,January and February (Fig. 3d); clearly illustrated, for example, byARNTL expression (Supplementary Fig. 6B). The pattern ofseasonal gene expression in samples from Iceland was unique(Supplementary Fig. 4).Seasonal gene expression was not altered in samples from

children with self-reported infections35 (Supplementary Fig. 5),and the type-I interferon response gene, SIGLEC135, was notseasonal. Finally, recruitment into the asthma cohort wasdependent on participants being free from infectious diseases34.Nevertheless, the relationship between seasonal infections anddiseases, and these seasonal gene expression patterns, remains tobe fully described.

Common seasonal genes in the immune system. One hundredand forty-seven genes showed common seasonality in theBABYDIET, T1D, Australia, USA and UK/Ireland datasets(Supplementary Table 9). These 147 genes had similar seasonalexpression patterns in each cohort (Supplementary Fig. 6).

ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms8000

2 NATURE COMMUNICATIONS | 6:7000 | DOI: 10.1038/ncomms8000 |www.nature.com/naturecommunications


Seasonality of circadian clock genes

Rel

ativ

e Lo

g2 (e

xpres

sion

)

Rel

ativ

e lo

g2 (e

xpres

sion)

Seasonality of ARNTL, VDR, B2M and GAPDH gene expression

1.041023

1.621006

9.181001

4.111001

Morning and afternoon clinics

Seasonal ARNTL expression isindependent of circadian phase sampling

Seasonality of hormone receptor genes

Rel

ative

lo

g2 (e

xpre

ssio

n)R

elat

ive lo

g2 (e

xpre

ssio

n)

Rel

ative

lo

g2 (e

xpre

ssio

n)

0.2

0.0

0.2

0.2

0.0

0.2

0.2

0.0

0.2

Jan

Feb

Mar Ap

r

May

June July

Aug

Sep

Oct

Nov

Dec

Jan

Feb

Mar Ap

rM

ayJu

ne July

Aug

Sep

Oct

Nov

Dec

Jan

Feb

Mar Ap

rM

ayJu

ne July

Aug

Sep

Oct

Nov

DecMonths

Jan

Feb

Mar Ap

rM

ayJu

ne July

Aug

Sep

Oct

Nov

Dec

Months

Jan

Feb

Mar Ap

rM

ayJu

ne July

Aug

Sep

Oct

Nov

Dec

Months

GeneARNTL

VDR

B2M

GAPDH

ARNTL, 7938563

0.2

0.1

0.0

0.1

0.2

GeneARNTL 1.0436e23

CLOCK 3.0810e08

TIM 1.5511e24

RORA 6.4048e18NR1D2 2.0893e08

NFIL3 4.4120e13

CRY1 7.3177e19CSNK1D 2.6668e23CSNK1E 2.4008e15

GeneADIPOR1 2.9972e10CUL5 2.7908e13

PTGIR 1.9316e10PTGER4 1.0836e05PTGDR 5.4521e19OXER1 2.5942e15

CYSLTR1 5.4995e21ESR2 7.8330e09NR3C1 5.0469e19

0.2

0.2

Months

Months

0.0

P=1.043618e23

P=1.398571e13

Blood samples collected during:morning clinics only

ARNTL, 7938563

Figure 1 | Seasonal mRNA expression in the peripheral human immune system. Relative expression proles of seasonal genes (tted values of the

cosinor model). (a) ARNTL expression was increased in the summer months of June, July and August (ANOVA, w22, P 1.04 10 23), compared with thewinter months of November through February (1.5097-fold difference between February and August (n 109 individuals). Similarly, the nuclear vitamin Dreceptor (VDR) shows peak expression in June through August (ANOVA, w22, P 1.62 1006). The housekeeping genes, B2M and GAPDH, did not haveseasonal expression proles. (b) Seasonal ARNTL expression in PBMCs independent of the circadian phase. Similar seasonal ARNTL expression proles

were observed regardless of whether blood samples were collected during morning (BABYDIET, n 109 individuals) or afternoon clinic visits (T1D cohort,n 236 individuals). (c) In the BABYDIET data set, nine known components of the circadian clock had seasonal expression proles in the peripheralimmune system, as did certain hormone, leukotriene and prostaglandin receptors. (d) The receptors for the anti-inammatory glucocorticoids (NR3C1) and

the pro-inammatory prostaglandins (PTGDR, PTGIR and PTGER4) and leukotrienes (CYSLTR1) had opposing seasonal expression proles.

NATURE COMMUNICATIONS | DOI: 10.1038/ncomms8000 ARTICLE



Notably, in the Icelandic cohort, the common seasonal genes didnot share the same expression pattern (Supplementary Fig. 4).This could be due to near-24-h daylight during summer if sea-sonal human physiology is regulated by changes in the annualphotoperiod36. ARNTL was found to be a common seasonal gene(Fishers method, w210, P 6.73 10 57), with increased summerexpression in each PBMC data set, except Iceland. The gene withthe strongest seasonal prole common to all data sets (excludingIceland) was C14orf159 (winter expressed, Fishers method,w210, P 3.93 10 66). The mitochondrial protein, UPF0317,encoded by C14orf159 (whose expression is regulated byoestrogen receptor alpha37) is highly conserved in chordates,although its function in humans is largely unknown.

Seasonal cellular remodelling of the human immune system. AsPBMCs represent several specialized haematopoietic lineages, we

sought to determine whether seasonal gene expression resultedfrom annual changes in the cellular composition of blood.In support of this, we found the expression of seasonal genes to

correlate strongly with the expression of 13 genes known to markdifferent immune cell types present in PBMCs38 (Fig. 4a).Furthermore, by analysing full blood count (FBC) data from7,343 healthy adult donors enroled in the Cambridge BioResource(United Kingdom), we found the total number of white bloodcells (ANOVA, F-test, P 1.75 10 10), lymphocytes (ANOVA,F-test, P 2.11 10 11), monocytes (ANOVA, F-test, P 9.14 10 30), basophils (ANOVA, F-test, P 2.74 10 6),eosinophils (ANOVA, F-test, P 0.00235), neutrophils(ANOVA, F-test, P 6.13 10 27) and platelets (ANOVA,F-test, P 2.02 10 12) to exhibit seasonality in the peripheralcirculation, as did the mean corpuscular volume (MCV)(ANOVA, F-test, P 1.32 10 21) and mean corpuscular

Seasonal genes in the BABYDIET data set

Genes defined as summer, winter or neitheramongst BABYDIET seasonal genes

5,000 genes not defined asseasonal

Rel

ativ

e lo

g2 (e

xpres

sion

)

2,311 summer genes2,826 winter genes

Rel

ative

lo

g2 (ex

pre

ssio

n)

Rel

ative

log2

(exp

ress

ion)

1.0

0.5

0.0

0.5

1.0

1.0

0.5

0.0

0.5

1.0

1.0

0.5

0.0

0.5

1.0

Jan

Feb

Mar Ap

rM

ayJu

ne July

Aug

Sep

Oct

Nov

Dec

Jan

Feb

Mar Ap

rM

ayJu

ne July

Aug

Sep

Oct

Nov

Dec

Months

MonthsJa

nFe

bM

ar Apr

May

June July

Aug

Sep

Oct

Nov

Dec

Months

Figure 2 | Widespread seasonal mRNA expression in the human immune system. Relative expression proles of seasonal genes (tted values of the

cosinor model). (a) A total of 5,136 genes (B23% of the protein-coding genome) were identied as having seasonal variation in expression (genome-widesignicance, Pr1.52 1006) in the BABYDIET PBMC data set. (b) Five thousand randomly selected genes not identied as seasonal are shownas a comparison. (c) Two anti-phasic patterns of gene expression were observed among seasonal genes. We dened the majority of seasonal genes as

being either winter- (green) or summer-expressed (blue). In BABYDIET, 2,311 genes were increased in expression in the summer and 2,826 were increased

in the winter. One of the seasonal probes did not fall into our denition of summer and winter, shown as a red line.




haemoglobin (MCH) (ANOVA, F-test, P 1.73 10 15) oferythrocytes (Fig. 4b). Our results are in agreement with a studythat reported seasonal red blood cell and platelet geneexpression39.In a more equatorial cohort, comprising 4,200 healthy

individuals from The Gambia (West Africa), we observedseasonal variation in the number of total white blood cells(F-test, P 0.011), lymphocytes (F-test, P 1.40 10 05),

monocytes (F-test, P 8.71 10 16) and platelets (F-test,P 2.07 10 18) (Fig. 4c), but not granulocytes. We alsoobserved striking seasonal variation in red blood cell numbers(F-test, P 8.43 10 30) and their mean corpuscular haemo-globin (F-test, P 4.07 10 30) (Supplementary Fig. 7). Theseasonal patterns in The Gambia were completely distinct tothose observed in the UK cohort. In The Gambian cohort, thenumbers of all seasonal cell types peaked during the rainy season

United Kingdom and Ireland Australia United States

United Kingdom and Ireland Australia United States

Seasonal genes in T1D PBMCs

791 genes 1,257 genes 409 genes

1,697 genes

Summer & winter (BABYDIET genes)in T1D PBMCs

Rel

ativ

e lo

g2 (e

xpres

sion)

Rel

ativ

e lo

g2 (e

xpres

sion)

Rel

ativ

e lo

g2 (e

xpres

sion)

Rel

ativ

e lo

g2 (e

xpres

sion)

Rel

ativ

e lo

g2 (e

xpres

sion)

Rel

ativ

e lo

g2 (e

xpres

sion)

Rel

ativ

e lo

g2 (e

xpres

sion)

Rel

ativ

e lo

g2 (e

xpres

sion)

Seasonal genes in different populations of adult asthmatic patients

Summer and winter genes from BABYDIET in the adult asthmatic patients

0.4

0.0

0.4

Jan

Feb

Mar Apr

May

June July

Sep

Oct

Nov

Dec

Aug

Jan

Feb

Mar Apr

May

June July

Sep

Oct

Nov

Dec

Aug

Jan

Feb

Mar Ap

rM

ayJu

ne July

Sep

Oct

Nov

Dec

Aug

Jan

Feb

Mar Ap

rM

ayJu

ne July

Sep

Oct

Nov

Dec

Aug

Jan

Feb

1

0 0

1

Months Months

Months MonthsMonths

Months Months

Months

11.0

0.5

0.0

0.5

1.0

Mar Ap

rM

ayJu

ne July

Sep

Oct

Nov

Dec

Aug

Jan

Feb

Mar Apr

May

June July

Sep

Oct

Nov

Dec

Aug

Jan

Feb

Mar Apr

May

June July

Sep

Oct

Nov

Dec

Aug

Jan

Feb

Mar Apr

May

June July

Sep

Oct

Nov

Dec

Aug

0.4

0.0

0.4

1

0

1 1

0

1

1.0

0.5

0.0

0.5

1.0

1

Figure 3 | Seasonal gene expression in geographically distinct cohorts. (a) Seasonality was also observed in PBMCs collected from T1D patients in the

United Kingdom (n 236 individuals). A total of 1,697 genes were seasonal in this data set. (b) The previously dened summer and winter genes from theBABYDIETdata set maintained their seasonal expression patterns in the T1D samples. (c) PBMCs from asthmatic patients collected from different countries also

showed seasonal gene expression. In the United Kingdom/Ireland (n 26 asthmatic individuals; 85 PBMC samples), 791 genes were seasonal, while 1,257 and409 genes were seasonal in Australia (n 26 individuals; 85 samples) and United States (n 37 individuals; 123 samples), respectively. (d) Summer and winterBABYDIET genes maintained their seasonal expression patterns in the asthmatic PBMC samples, with their patterns inverted in Australia.




Rel

ativ

e ce

ll fre

quen

cy u

nder

the

cosin

or m

odel

White blood cells Lymphocytes Granulocytes

Monocytes Platelets Mean platelet volume

P =0.011 P =1.401005 P =0.151

P =8.711016 P =2.071018 P =6.721015

Seasonal variation in cellular composition of blood samples from the Gambian equatorial cohort

Fitted valuesIndividual data points

Stan

dard

uni

ts fo

r eac

h ha

emat

olog

ical p

aram

eter

DCs

CD4+ T

ActivatedCD4+ T

B cells

ActivatedNKs/DCs

ActivatedNK cells

NK cells

Monocytes/Neutrophils

Activated DC

Activated B cells

Macrophages

Neutrophils

P=2.7410

06

P=0.00235

NS

P=2.11x10

11

P=2.1310

05

P=1.73x10

15

P=1.32x10

21

P=9.14x10

30

P=1.75x10

10

P=2.17x10

05

P=2.02x10

12

P=6.13x10

27

Seasonal variation in cellular composition of blood samples fromhealthy volunteers, Cambridge BioResource, UK

BABYDIET seasonal genes correlate in expressionwith genes identifying unique immune lineages

CD4+ T cells

01

1

234

234

0

0123401234

Den

sity

012340123401234

0.

5

0.0

Spearman correlation

0.5

1.0

0.

5

0.0

0.5

1.0

0.

5

0.0

Spearman correlation

0.5

1.0

0.

5

0.0

0.5

1.0

Non-Seasonal Seasonal Non-Seasonal Seasonal

CLIC3FCN1

PDCD1LG2

PEG10

SPARCTN

FRSF10C

CD1ECRIP2

CTLA4M

S4A1RSAD2

SBK1SRRM

2

Den

sity

01234

01234

01234

01234

01234

01234

0.004

0.000

0.004

0.10

0.05

0.00

0.05

0.10

0.2

0.1

0.0

0.1

0.2

JanFebM

arAprM

ayJuneJulyAugSepO

ctN

ovD

ec

JanFebM

arAprM

ayJuneJulyAugSepO

ctN

ovD

ec

JanFebM

arAprM

ayJuneJulyAugSepO

ctN

ovD

ec

JanFebM

arAprM

ayJuneJulyAugSepO

ctN

ovD

ec

0.01

0.00

0.01

0.2

0.1

0.0

0.1

0.2

0.1

0.0

0.1

0.25

0.00

0.25

1.0

0.5

0.0

0.5

1.0

0.04

0.02

0.00

0.02

0.04

5

0

5

0.06

0.03

0.00

0.03

0.06

0.2

0.1

0.0

0.1

0.2

BAS_Count EOS_Count HCT HGB

LYM_Count MCH MCV Mon_Count

Neu_Count PLT RBC WBC

Months

25

20

15

10

5

0

0 2 4 6 8 10 12 0

0

2

4

6

8

1015

10

5

0

12

10

8

60

200

400

600

800

10001.5

1

0.5

0

2 4 6 8 10 12MonthsMonths

0 2 4 6 8 10 12Months

0 2 4 6 8 10 12Months

0 2 4 6 8 10 12Months

0 2 4 6 8 10 12Months

Figure 4 | Seasonal changes in the cellular composition of human peripheral blood. (a) Expression levels of 13 genes that have been used to identify

different blood cell types among total PBMCs were strongly correlated (positively and negatively) with seasonal genes identied in the BABYDIETdata set.

In comparison, non-seasonal genes were less correlated with these marker genes, although exceptions exist: CTLA-4 expression also correlations with

non-seasonal genes. (b) Indeed, by analysing full blood count data obtained from 7,343 healthy adult donors enroled in the Cambridge BioResource, we

found the cellular composition, and other haematological parameters of blood to vary by season. HCTwas the only response that did not show seasonal

variation. (c) Distinct seasonal variation in cell counts was observed in a cohort of 4,200 healthy adults and children from The Gambia. EOS, eosinophils;

LYM, lymphocytes, NEU, neutrophils, PLT, platelets; RBC, red blood cells; WBC, total white blood cells; BAS, basophils; HGB, haemoglobin; MCH, mean

corpuscular haemoglobin; MCV, mean corpuscular volume; MON, monocytes; HCT, haematocrit.




(as previously reported for leukocytes40), June through October,during which time the immune system faces different pathogenicchallenges, such as an increased infectious disease burden,including malaria41.

Seasonal differences in human immunity. To address whetherimmunological function varies seasonally, as suggested by thetranscriptomic and cell count data, we generated modules ofco-regulated seasonal mRNAs identied in the BABYDIET dataset (Fig. 5a and Supplementary Fig. 8 and SupplementaryTable 12). Among the seven winter-expressed modules weidentied, we found pro-inammatory processes to be morefrequent, compared with the identied summer-expressed mod-ules. B-cell receptor (BCR) signalling (Hypergeometric test,P 3.39 10 10), FcR-gamma-associated processes (Hypergeo-metric test, P 4.45 10 6), lysosomes (Hypergeometrictest, P 2.96 10 5), chemokine signalling (Hypergeometrictest, P 3.56 10 5) and phagosomes (Hypergeometric test,P 5.97 10 5) were all strongly associated with winter-expressed modules. In contrast, RNA transport (Hypergeometrictest, P 1.70 10 7), RNA degradation (Hypergeometric test,P 1.02x10 5), ubiquitin-mediated proteolysis (Hypergeometrictest, P 0.0002), circadian rhythms in mammals (Hypergeo-metric test, P 0.0011) and splicosome (Hypergeometric test,P 0.0014) were the most-associated pathways with summer-expressed modules, suggesting that a more inammatory status ofthe immune system predominates in winter (Fig. 5b andSupplementary Fig. 8).In further support of this, we found the concentration of

sIL-6R protein to be increased in winter in samples fromBABYDIET and BABYDIAB (a related collection42; ANOVA, w22,P 2.74 10 11), in complete agreement with the increasedwinter expression of IL6R mRNA in BABYDIET samples(ANOVA, w22, P 9.33x10 12) (Fig. 5c). sIL-6R is an importantorchestrator of leukocyte recruitment43 and trans-presents IL-6 tocells expressing gp130 in the absence of the cell-surface IL-6R44,endowing IL-6 with a broader spectrum of inuence. Indeed, acoding variant in IL6R that alters circulating sIL-6Rconcentration is associated with impaired IL-6 signalling andthe protection from cardiovascular disease, rheumatoid arthritisand T1D45. Interestingly, early-stage inammation in rheumatoidarthritis (a disease treated with anti-IL-6 receptor reagents46) hasbeen shown to either resolve or progress to erosive disease, and apredictor of this outcome is the season when disease symptomsrst present47. T1D also has seasonal trends in diagnoses9 andautoantibody positivity48, suggesting that seasonal environmentsimpact on autoimmune disease pathologies. Furthermore, wefound the circulating level of the acute-phase complementactivator, C-reactive protein49, to be increased during wintermonths (Fig. 5d).These gene expression data also suggest that the quality of a

vaccine response may be inuenced by season. We foundthe expression of TLR7 (ANOVA, w22, P 4.22 10 16),TLR8 (ANOVA, w22, P 3.70 10 09) and DDX58 (encodingthe viral RNA receptor RIG-I, ANOVA, w22, P 9.65 10 20)to have increased in expression in winter months in theBABYDIET data set: the increased expression of these genescorrelated with protective immunity in response to the YellowFever vaccine (YF-17D)50. TNFRSF17 also showed seasonalvariation in the BABYDIET data set (ANOVA, w22,P 1.30 10 12), and its induction is shown to be predictiveof an antibody response after trivalent inuenza vaccine51,52.Furthermore, OAS1 (ANOVA, w22, P 1.43 10 20), OAS2(ANOVA, w22, P 2.85 10 16), STAT2 (ANOVA, w22, P 1.76 10 18), POU2AF1 (ANOVA, w22, P 3.27 10 15) and

CD27 (ANOVA, w22, P 3.29 10 16) were also seasonaland their expression in PBMCs was correlated with increasedanti-DT IgG responses after the meningococcal vaccine(MCV4)51. The antibody responses to rabies, typhoid andpneumococcal vaccines are inuenced by the month of vaccineadministration40.

Seasonal gene expression in subcutaneous adipose tissue. Giventhe remarkable seasonality of the peripheral immune system andthe correlations we found with multiple health-associated phe-notypes, we anticipated that tissues throughout the body woulddisplay extensive seasonality of gene expression. We were able toanalyse gene expression data from a collection of subcutaneousadipose tissue samples obtained from 825 healthy female donorsenroled in the TwinsUK cohort53.We found 4,027 genes to be seasonally expressed (Fig. 6,

Supplementary Table 13), including IL6ST (gp130) (ANOVA,w22, P 2.55 10 8) and IL6R (ANOVA, w22, P 1.49 10 8):adipose tissue can produce IL-654. One thousand, two hundredand thirteen genes were common to both adipose tissue andBABYDIET data sets (Supplementary Table 14), suggesting thatcommon genetic mechanisms regulate seasonality.In adipose tissue, as in PBMCs, metabolic pathways were

among the most associated seasonal pathways (SupplementaryFig. 9). Such seasonal metabolic programmes may have beenselected for due to annual differences in temperature anddiet. Adipose tissue seasonality has important implications forimmunology, obesity and metabolic disease research; for example,PPARG, targeted by thiazolidinediones as a current treatment oftype 2 diabetes, was found to be seasonal in adipose tissue(ANOVA, w22, P 3.75 10 9).

DiscussionEcological changes alter the types and dynamics of inter- andintra-organism biological processes, and it follows that suchchanges will be manifested as seasonal transcriptional signatureswithin the immune systems of different organisms, which adaptto their environment. Studies of the function, dynamics andvariability of the immune system are undergoing a long-awaitedrenaissance partly owing to the development and application ofnew phenotyping technologies55,56. Nevertheless, to date, nostudy to our knowledge has taken into account the variability wehave observed in the immune system according to season, whichcould, for example, increase the differences in some immunephenotypes between twins or other family members if bloodsamples were collected at different times of year. We observedseasonal differences in expression across a large number of genesin mixed populations of human peripheral white blood cells fromgeographically and ethnically diverse locations, and, remarkably,seasonal genes displayed opposing patterns in the Southern andNorthern hemispheres. Fewer seasonal genes were identied inIcelandic donors, and common seasonal genes had a less similarseasonal pattern in this data set. If a seasonal photoperiodic clockexists in humans, the impact of living at higher latitudes requiresfurther exploration.These periodically changing transcriptional landscapes in

PBMCs, which appear to be predominantly driven by annualchanges in the cellular composition of blood, are likely toinuence various aspects of the human immune response. Indeed,the increased winter expression of co-regulated pro-inammatorygene modules, the functionally important increased concentrationof sIL-6R and CRP in the blood, and the observation that a loss ofBMAL1 (ARNLT was reduced in winter) promotes inammationin mice28, strongly suggests that the immune system is morepro-inammatory in Europeans during the northern hemisphere




Rel

ativ

e lo

g2 (e

xpres

sion)

sIL-6R protein in BABYDIET/DIAB

Two

co-re

gula

ted

gene

mod

ules

Rel

ativ

e lo

g2 (e

xpres

sion)

Rel

ativ

e Lo

g2 (e

xpres

sion)

Summer Winter

RNA processing

Seasonal circulating CRP levels

Rel

ativ

e lo

g (H

SCRP

)co

nce

ntra

tion

leve

ls (m

g/l)

0.7

0.6

0.5

0.4

0.3

0.2

0.1

0.0

Hei

ght

Gre

enye

llow

Gre

en

Mag

enta

Purp

le

Blac

k

Brow

n

Red

Turq

uoise

Yello

w

Blue

Pink

Gre

y

Clustering tree based of the module eigengenes0.4

0.0

0.4

0.2

0.0

0.2

Jan

Feb

Mar Ap

rM

ayJu

ne July

Aug

Sep

Oct

Nov

Dec

Months

Months

P =9.332427e120.2

0.1

0.0

0.1

0.2

Jan

Feb

Mar Ap

rM

ayJu

ne July

Aug

Sep

Oct

Nov

Dec Jan

Feb

Mar Ap

rM

ayJu

ne July

Aug

Sep

Oct

Nov

Dec

MonthsMonths

2

0

2

0.04

0.00

0.04

Jan

Feb

Mar Apr

May

June July

Aug

Sep

Oct

Nov

Dec

Months

As. dist(dissimME)hclust (*, ''average'')

Jan

Feb

Mar Apr

May

June July

Aug

Sep

Oct

Nov

Dec

Remainingseasonal genesnot in modules

Associated pathways:BCR receptor signallingmetabolic pathways

Phagosomelysosomebacterial infections

B-cell receptor signalling

CR2FCGR2BCD72BLNK2CD79BBTK

RANBP2EIF3JRAE1NUP54DDX20STRAPNUPL1PAIP1

RNA transportP =1.701007RNA degradationP =1.021005

P =2.74x1011

Rel

ativ

e sI

L-6R

con

cent

ratio

n (ng

/ml)

P =5.811006

IL6R mRNA in BABYDIET

P =3.391010

Figure 5 | Inammatory responses predominate the immune system in Europe. (a) Co-regulated seasonal gene modules were generated

to analyse differences in immune function by season: eight winter modules and three summer modules were generated. (b) Two modules of seasonally

co-regulated genes from the BABYDIETdata set are shown as examples. A module consisting of genes involved in B-cell receptor signalling, (including CR2,

BLNK, BTK, FCGR2B, CD72, CD79B) was more highly expressed in the winter, as was a module associated with metabolic processes. In contrast,

a RNA-processing module (containing RANBP2, EIF3J, RAE1, NUP54, DDX20, STRAP, NUPL1, PAIP1) was more highly expressed in the summer. (c) IL6R

mRNA expression was increased in the winter, in BABYDIET samples (ANOVA, w22, P 9.33 10 12), as was observed for the circulating level of sIL-6Rprotein in the serum of BABYDIET/DIAB children (ANOVA, w22, P 2.74 10 11). (d) The circulating levels of C-reactive protein displayed seasonalvariation in a cohort of 3,412 donors diagnosed as hypertensive but not conventionally dyslipidemic. ASCOT enrolled participants in Ireland, Denmark,

Finland, Iceland, Norway, Sweden and the UK (two measurements per donor), with increased levels present during winter HSCRP - high sensitivity

C-reactive protein.




winter. In mice, Arntl-BMAL1 controls the diurnal variationof circulating and tissue-resident inammatory monocytenumbers28, although how ARNTL controls human immunefunction is not known. We note that, in Europeans, totalmonocyte numbers in blood are increased during winter, whenARNTL expression is the lowest. Notably, acute-phase proteinsincluding CRP are induced by IL-6, which can be produced bymacrophages and adipocytes54. This entire network could be amajor factor in the higher frequency of cardiovascular disease-associated deaths in winter6, when increased risk is associatedwith excessive inammation, IL-6 and monocytes. Furthermore,increased IL-6 signalling is associated with increased risk ofrheumatoid arthritis and type 1 diabetes45, which peaks inincidence during the European winter. Increased IL-6 signallingand elevated CRP levels have also been associated withneuropsychiatric symptoms in children and adults57,58. Thus,modulation of IL-6 signalling according to season could beconsidered as a therapeutic strategy in various disease contexts.Whether a seasonal human immune system contributes to host-mediated pathology and morbidity after infection59 remains to bedetermined, but the correlations we report suggest this might bethe case.The breadth and functional characteristics of the seasonal gene

expression we observed suggest that it has been evolutionarilyselected for. During European winters, the thresholds required totrigger an immune response may be lower as a direct consequenceof our co-evolution with infectious organisms and increasedinter-species competition during winter, especially as humansmigrated out of Africa to colder, more seasonally pronouncedlatitudes. In our European cohorts, winter was associated withincreased monocytes and inammation, while FBC data from themore-equatorial Gambian cohort exhibited distinct seasonalvariation in cell numbers. In this data set, seasonal peaks in cellnumbers correlated with the rainy season (June to October),

during which time the infectious disease burden is at its highestlevels.Regardless of any particular causal factor driving these

differences, which are likely many, our results demonstrate thatdifferent human populations independently vary the cellularcomposition of their immune system by season, suggestive ofdistinct environmental adaptations. Furthermore, although ourdata suggest that cell-type numbers contribute the majority ofseasonal gene expression in PBMCs, future studies of seasonalphenotypic differences within puried immune cell subsets arelikely to reveal an additional layer of complexity in the humanimmune system.The origin and likely diverse mechanisms maintaining seasonal

variation remain to be established: daylight and ambienttemperature are candidate environmental cues that could co-ordinate seasonal hormonal phenotypes and cell-fate decisions inhaematopoietic and stem cells. Indeed, diurnal entrainment of thehuman circadian clock requires daylight changes, demonstratingthat humans sense and process photoperiodic cues to co-ordinatephysiology.The environmental perturbation of our molecular clocks is

thought to be deleterious to health60, which may help explainingthe increasing complex disease burden in industrializedcountries61 and populations at extreme latitudes9, where clockdysregulation or chronodisruption may be more frequent62. Inseasonally-breeding mammals, circadian melatonin productioncues reproduction in response to changes in the annualphotoperiod63. In the arctic mammal, Rangifer tarandus, dailymelatonin rhythms are acutely responsive to the night-day phasebut not the circadian phase64, demonstrating species-specicadaptation to the unique night-day cycles present at extremelatitudes: the ability of humans to properly function in suchenvironments is not well understood. Furthermore, a circannualmolecular clock was recently shown to control seasonalreproduction in hamsters, independently of melatonin and sexsteroids, yet using the same neuroendocrine reproductivepathway65. Human genetic variation in the ARNTL gene regionhas been associated with age of menarche66,67, which is alsoseasonal.The widespread seasonal gene expression observed in sub-

cutaneous adipose demonstrates seasonality across differenthuman tissues.Regardless of the mechanisms causing and maintaining these

and other seasonal variations, our results provide a plausiblemechanism to explain part of the seasonality of human disease.These data provide a fundamental shift in how we conceptualizeimmunity in humans, and we propose that seasonal changes bemore broadly considered as major determinants of humanphysiology.

MethodsStudy subjects and human samples. All samples and information were collectedwith written and signed informed consent. One hundred and nine childrengenetically predisposed to T1D were enrolled in the BABYDIET study. TheBABYDIET study is an intensively monitored dietary intervention study testing thepotential effect of delayed gluten exposure on the development of islet auto-immunity in children at increased risk for diabetes in Germany. Children youngerthan 3 months with at least one rst-degree relative with T1D and one of threespecic T1D-associated HLA genotypes (DRB1*03-DQA1*05:01-DQB1*0201/DRB1*04-DQA1*03:01-DQB1*03:02; DRB1*04-DQA1*03:01-DQB1*03:02/DRB1*04-DQA1*03:01-DQB1*03:02 or DRB1*03-QA1*05:01DQB1*02:01/DRB1*03-DQA1*05:01DQB1*02:01) were recruited between 2,000 and 2,006(participation rate: 88.8 %) and randomized to exposure to dietary gluten fromage 6 months or from age 12 months. After inclusion, children were followed inthree monthly intervals until the age of 3 years and yearly thereafter for efcacy(persistent islet autoantibodies) and safety assessment, including intensive mon-itoring with three monthly sample collection of venous blood, urine and stool.PBMCs were isolated from venous blood samples taken at each visit and storedat 80 C in TRIZOL.

Seasonal gene expressionin adult subcutaneous adipose tissue

0.50

0.25

0.00

0.25

0.50Jan Feb Mar Apr May June July Aug

MonthsSep Oct Nov Dec

4,027 Genes

Rel

ativ

e lo

g2 (e

xpres

sion)

Figure 6 | Seasonal gene expression in subcutaneous adipose tissue. In a

collection of 856 female adult donors from the United Kingdom, 4,027

genes were found to be seasonal in adipose tissue. As observed in PBMCs,

two distinct anti-phasic proles were present.




The T1D PBMCs were collected as part of the Genetic Resource InvestigatingDiabetes (GRID)cohort collection (http://www.childhood-diabetes.org.uk/grid.shtml) by our laboratory and others. Blood samples were collected in morning orafternoon hospital clinics, across more than 150 centres in the United Kingdom.Blood was collected into ACD vacutainers and PBMCs isolated separatedusing either Sigma Accuspin or Histopaque according to the manufacturersrecommendations. PBMCs were cryopreserved in the presence of DMSO andstored at 80 C until use. For RNA isolation, PBMC samples were thawed,washed with X-Vivo 15 (Lonza) and added to Trizol reagent. RNA was isolated andgene expression data were generated in the same way as the BABYDIET cohort,described in the following section.

All BABYDIET and T1D PBMC gene expression data are deposited withArrayExpress (accession number: E-MTAB-1724).

Gene expression data for the multi-centre asthma cohort is publically available.The cohort was collected and processed as described by Bjornsdottir et al.34,and the raw and normalized data are deposited with ArrayExpress (http://www.ebi.ac.uk/arrayexpress/, E-GEOD-19301). This gene expression data set wasgenerated on Affymetrix HG-U133A GeneChip Array, which we found to have22,283 probesets (which map to 10,457 unique ENSEMBL gene identiers).The inclusion of patients in the initial collection of the study was dependent onparticipants being free from active infection, major intercurrent illness, allergenimmunotherapy, pregnancy and lactation34.

The available processed data as well as the R ExpressionSet le weredownloaded from ArrayExpress. Information regarding the disease phase of thesamples, their country of origin and the date of bleeding was used in our analyses.Only asthma patients dened as being in a quiet disease phase were included in ouranalyses. Precise age at bleed for each donor was also not available in this data set,although individuals between 18 and 83 years of age were present in the cohort.The mean age of the asthma patients was 45.08 years34.

As information regarding sample gender in this data set was not available, wedened gender based on Y-expressed genes in PBMCs (DDX3Y, KDM5D, USP9Y,and RPS4Y1). The rst principal component of the expression of the listed geneswas calculated for each individual in the study. Patients with component valuessmaller than zero were classied as female and patients with component valuesgreater than zero were classied as male.

The asthma PBMC data set was divided into four groups according to countryof sample collection; United States, Australia, United Kingdom/Ireland andIceland.

The subcutaneous adipose tissue gene expression data were collected by theMuTHER consortium53 and is publically available (Array Express E-TABM-1140).The adipose tissue data set includes 825 female twins, among them 80 singletons,448 dizygotic and 297 monozygotic individuals. Gene expression data for 48,638probesets (mapping to 24,332 unique Entrez genes) were downloaded.

Sample numbers included in our analyses of each cohort, and their monthlydistributions, are shown in Supplementary Table 1.

Gene expression analysis in BABYDIET and T1D PBMCs. Only gene expressiondata for the BABYDIET and T1D PBMC cohorts were generated in our laboratory.In brief, single-stranded cDNA was synthesized from 200 ng total RNA using theAmbion whole-transcript expression kit (Ambion) according to the manufacturersrecommendations. A total of 3.44 mg cDNA was fragmented and labelled using theGeneChip terminal labelling and hybridization kit and hybridized to 96-sampleTitan Affymetrix Human Gene 1.1 ST arrays, which provide comprehensive whole-transcriptome coverage. After quality control, we measured the expression of33,297 probesets, which map to 22,822 unique ENSEMBL gene identiers.

BABYDIET, T1D and adipose gene expression data were summarized byexon-level probesets and normalized using variance stabilizing normalization:post quality control 454 BABYDIET35, 236 T1D and 825 adipose samples wereused for analysing gene expression.

The gene expression data of the asthmatic patients were log2 transformedbefore any analysis.

Climatic data for modelling seasonal gene expression. Historical raw data forthe mean daily temperature, as well as the total daily hours of sunlight in Munich(Germany), were obtained from the Integrated Climate Data Centre at theUniversity of Hamburg (http://icdc.zmaw.de/dwd_station.html?&L=1).

For the analysis of the T1D PBMC data that came from all around UnitedKingdom we downloaded the maximum and minimum temperature data fromseven stations across United Kingdom (Armagh, Camborne, Eskdalemuir, Lerwick,Stornoway airport and Valley) from the National Climatic Data Centre, USA(http://www.ncdc.noaa.gov/cdo-web/search) and averaged readings across allstations.

For the analysis of the asthma cohort (ArrayExpress: E-GEOD-19301), the dailymaximum and minimum temperature for relevant cities/regions in the UnitedKingdom (Central England UK station at Birmingham), United States (New Jersey,Seattle, Atlanta, New Haven), Iceland (Reykjavik), Ireland (Dublin) and Australia(Melbourne, Perth, Adelaide) were obtained from the National Climatic DataCentre, USA and The Digital Technology Group. The average temperature valueswere computed and used in subsequent analyses.

Self-reported infections in BABYDIET cohort. At each visit, parents ofBABYDIET children completed a detailed questionnaire on their childrens historyof infections, fever and medication. Specically, they were asked about fever,infectious symptoms (such as diarrhoea, vomiting, constipation and allergies) andthe name of administered pharmaceutical agents or their active ingredient withstarting date and duration of infections and medication. Infectious disease wasdened as an acute event according to the ICD710 Code or by a symptomindicating an infectious genesis. Infectious events were assigned to a specic timeinterval by their date of onset, and infectious events that could be matched tomicroarray samples were included for analysis, as described35. Other disease eventssuch as allergies or accidents were not considered as infectious diseases.

Soluble IL-6 receptor ELISA. Circulating sIL-6R concentrations were measured inBABYDIET and BABYDIAB serum samples using a highly sensitive non-isotopictime-resolved uorescence ELISA assay based on the dissociation-enhancedlanthanide uorescent immunoassay technology (DELFIA; PerkinElmer), asdescribed45. Test samples were diluted 1:20 in PBS 10% FBS and measured induplicate on 384-well MaxiSorp microtiter plates (Nunc), coated with 1 mgml 1monoclonal anti-human IL-6R antibody (clone 17506; RD Systems). Detection wasperformed using a biotinylated mouse anti-CD126 monoclonal antibody (cloneM182, BD Biosciences) diluted to a nal concentration of 100 ngml 1 inPBS 10% FBS and a Europium-Streptavidin detection solution (PerkinElmer),diluted in PBS 0.05% tween, 1% BSA, 7 mgml 1 DTPA to a nal concentrationof 0.05 mgml 1. Quantication of test samples was obtained by tting the readingsto a human recombinant IL-6Ra (RD systems) serial dilution standard curve platedin quadruplicate on each plate. Data for 782 unique individuals existed from722 families.

Cambridge BioResource full blood count data (UK cohort). Full blood countdata were obtained from the Cambridge BioResource. BioResource volunteers aresubjected to a full blood count on the day of blood sample collection usingBeckman Coulter LH700, Beckman Coulter DXH800 5 part diff analyser or aSysmex 5 part diff analyser. The available months of bleed were from February toNovember (no FBC data was available for December) and took the numeric values2 to 11, respectively. Responses measured included counts for basophils, eosino-phils, lymphocytes, monocytes, neutrophils, platelets, erythrocytes and total whiteblood cells. HCT (haematocrit), HGB (haemoglobin concentration), MCH (meancorpuscular haemoglobin) and MCV (mean corpuscular volume) were alsoanalysed.

Full blood count data from The Gambia. The Gambian cohort was collected aspart of the Keneba Biobank (http://www.ing.mrc.ac.uk/research_areas/the_kene-ba_biobank.aspx). All participants were recruited between 2012 and 2014 in theWest Kiang district and within the catchment area of the MRC InternationalNutrition Groups eld station at MRC Keneba. Supplementary Figure 7 givessummary statistics for the cohort. Written informed consent was obtained from allparticipants and all procedures were approved by the joint Gambian Government/MRC Ethics Committee. FBCs were available from 4,200 healthy individuals (at thetime of sample collection; 44.07% male) using a Medonic M-series analyser, whichmeasures the numbers of white blood cells, lymphocytes, granulocytes, monocytes,platelets and RBCs. Furthermore, it also analyses the mean platelet volume, RBChaemoglobin concentration, the haematocrit, MCV and MCH.

C-reactive protein. The level of CRP in the peripheral circulation was measured in3,412 donors (two samples per donor) collected as part of the ASCOT study68.Treatment with Atorvastatin did not remove the seasonal variation in thisparameter. Age and sex were included as covariates, while a random intercept wasadded for the individual identiers.

Statistical analysis of the data sets. Cosinor models with a period of 1 year weretted to test the effect of season on gene expression. The general formula of thetted model is given by:

Yjik a b cos 2ptik c sin 2ptik d fixed covariates grandom intercepts ejik

1

where Yjik represents the log2 expression of gene j for individual i recorded at timetik, with tik computed as the calendar day of the date of bleed divided by the totalnumber of days within the equivalent year.

The xed covariates and random intercepts terms were data-set-specic. For theanalysis of the BABYDIET and T1D data sets we added age at bleed and gender asxed effects covariates, whereas only gender was added as a covariate in theanalysis of the asthma PBMC microarray dataset (age was not available). Theidentity of each subject of the BABYDIET and of the asthma data sets weremodelled as a random intercept in the corresponding models. For the adiposetissue data set we modelled age at bleed as a xed covariate and added familyidentity and an indicator whether the twin was monozygotic or dyzogitic asrandom intercepts. Gender and age at bleed were treated as xed effects covariatesin the analysis of the soluble IL-6 receptor data, and family identity was included as




a random intercept. As only the month of bleed was available in the CambridgeBioResource FBC data, we adjusted the cosinor model to depend on month insteadof day; no other covariates were available and random intercepts were not required,as no individual was observed more than once. For the last two data sets, theresponse variable (Y) corresponds to IL-6R and to the tested FBC responses listedin the description of the data set. For analysis of CRP, age, sex and an age*sexinteraction were included as xed covariates, CRP was log transformed to removeright skew, and a random intercept was used to adjust for within individualrepeated measures.

To examine whether the effect of season was signicant we compared the ttedmodel in equation (1) with a model that did not include the effect of season. Thisalternative model is expressed by

Yjik a d fixed covariates g random intercepts ejik 2The P-value for season was determined by comparing the two models for eachgene using an analysis of variance test. Seasonal genes were classied as thosewith P values less than the data-set-specic Bonferroni correction thresholdalpha 0.05. For the BABYDIET and T1D data sets, we dened as seasonal thegenes with P values less than the corresponding Bonferroni correction P value andwith mean log2 expression greater than or equal to, 6.

The relative estimated log2 expression of each seasonal gene for each data setwas computed as

Y^ik b^ cos 2ptik c^ sin 2ptik 3where b^ and c^ are the least squares estimates of b and c of the model inequation (1), respectively.

Furthermore, we tested whether temperature or sunlight hours could predictgene expression of the PBMC data sets. Temperature and sunlight were dened,respectively, as the average temperature and number of sunlight hours over theweek preceding the date of bleed for each individual. For example the temperaturemodel is given by

yjik a b temperatureik d fixed covariates g random intercepts ejik4

The three alternative models for the seasonal cosinor function, sunlight andtemperature, each including only one of these predictors were tted to log2expression level for seasonal genes, as identied in each data set.

Denition of winter and summer seasonal genes in BABYDIET. Seasonal geneswere classied as winter genes if the relative estimated log2 expression values of thegenes were positive for all days of January, February and December and negativefor all days of June, July and August. In contrast, summer seasonal genes weredened as those with positive relative estimated log2 expression for all days of June,July and August and negative for all days of January, February and December.The fold change for each summer and winter gene was computed as two raised tothe power of the absolute difference of the estimated log2 expression between15 January and 15 July (days 15 and 196 of a 365-day calendar year).

Network and functional analysis of the seasonal genes identied in BABYDIET.A weighted co-expression gene network of the seasonal genes identied inBABYDIET was constructed using the R package WGCNA69. For the constructionof the network, individuals who sero-converted to T1D autoantibodies at any stageduring the BABYDIET study were not included. A scale-free topology network wascreated based on the seasonal genes, where the correlation of their log2 geneexpression was used as a measure of co-expression. Modules of highly correlatedgenes were detected through hierarchical clustering. Some genes were notcorrelated with other seasonal genes. The biological function of each module wasexamined through an over-representation pathway analysis carried out using theWEB-based GEne SeT AnaLysis Toolkit (WebGestalt, http://bioinfo.vanderbilt.edu/webgestalt/)70. The gene members of each module wereuploaded to WebGestalt and tested for over-representation within KEGGpathways. Pathways with less than three genes within our gene lists were excluded.The Hypergeometric test was applied, and theP values of the test were corrected for multiple testing using the BenjaminiHochberg method.

Analysis of the self-reported infections data of BABYDIET cohort. TheBABYDIET samples were divided into two categories, one that included all sampleswith no self-reported infections (57 samples) and one with all the samples with atleast one reported infection (152 samples). A principal component analysis (PCA)was performed, and the rst principal component from the analysis was used tosummarize the gene expression of BABYDIET seasonal genes. Similarly, the geneexpression of the genes within the black module (detected using network analysisof the seasonal genes identied in BABYDIET) was also summarized as the rstcomponent of a second PCA. The effect of infection on either of the two com-ponents was tested using analysis of variance. The black module was chosen as itcontained genes associated with the response to Staphylococcus infection.

Identication of common seasonal genes. We wanted to explore whether any ofthe seasonal genes identied in the PBMC cohorts were shared between the vedata sets (excluding Iceland). We compared the Bayesian information criterion(BIC) of the cosinor model (1) with the BIC of the model excluding the seasonalityeffect (2) for each of the genes from the two in-house data sets (BABYDIET andT1D) that had a P value o0.05/33297 in at least one of the two data sets. Thecommon seasonal genes of the two in-house datasets were dened as genes whoseBIC was smaller for (1) than (2) within each data set. We repeated the afore-mentioned steps to identify common seasonal genes in the asthma cohort. Theintersection of the two lists from the ve data sets were dened as commonseasonal genes.

We further computed a combined P value for the association of each commonseasonal gene by combining the P values of the ve data sets using Fishers productP value method.

Common seasonal genes between the adipose tissue data set and theBABYDIET data set were dened as the genes that were found seasonal for bothdata sets.

Seasonal analysis of The Gambian full blood count data. Given the differentseasonal climates present in West Africa compared to Europe, FBC parametersfrom The Gambia cohort were assessed through linear models that included sex,age (modelled through splines) and with seasonality modelled using three Fourierterms using STATA12.1. The signicance of season was assessed using an F-test.

Note added in proof. Adaptive oscillations at balanced polymorphisms inDrosophila in response to acute and persistent changes in climate were reportedwhile this work was under consideration (Bergland, A.O., Behrman, E.L.,OBrien, K.R., Schmidt P.S. & Petrov D.A. Plos Genet. 10(11):e1004775 (2014)).Furthermore, seasonally-variable associations of three genes involved in glucosemetabolism and circadian clock regulation, CRY1 (cryoptochrome 1), CRY2(cryoptochrome 1) and MTNR1B (melatonin receptor 1B) have recently beenreported in humans. (Renstrom, F., Koivula, R.W., Varga, T.V., Hallmans, G.,Mulder, H., Florez, J.C., Hu, F.B. & Franks, P.W. Diabetologia 10.1007/s00125-015-3533-8 (2015)).

References1. Kohler, M., Marin-Moratalla, N., Jordana, X. & Aanes, R. Seasonal bone growth

and physiology in endotherms shed light on dinosaur physiology. Nature 487,358361 (2012).

2. Yanovsky, M. J. & Kay, S. A. Molecular basis of seasonal time measurement inArabidopsis. Nature 419, 308312 (2002).

3. Ebling, F. J. On the value of seasonal mammals for identifying mechanismsunderlying the control of food intake and body weight. Horm. Behav. 66, 5665(2014).

4. A, A., Mohamed, E. R. & Ridi, R. E. Seasonal conditions determine themanner of skin rejection in reptiles. J. Exp. Zool. 265, 459468 (1993).

5. Mendoza, S. P., Lowe, E. L., Resko, J. A. & Levine, S. Seasonal variations ingonadal hormones and social behavior in squirrel monkeys. Physiol. Behav. 20,515522 (1978).

6. Pell, J. P. & Cobbe, S. M. Seasonal variations in coronary heart disease. QJM 92,689696 (1999).

7. Pell, J. P., Sirel, J., Marsden, A. K. & Cobbe, S. M. Seasonal variations in out ofhospital cardiopulmonary arrest. Heart 82, 680683 (1999).

8. Iikuni, N. et al. Whats in season for rheumatoid arthritis patients? Seasonaluctuations in disease activity. Rheumatology 46, 846848 (2007).

9. Moltchanova, E. V., Schreier, N., Lammi, N. & Karvonen, M. Seasonal variationof diagnosis of Type 1 diabetes mellitus in children worldwide. Diabet. Med. 26,673678 (2009).

10. Rosenthal, N. E. et al. Seasonal affective disorder. A description of thesyndrome and preliminary ndings with light therapy. Arch. Gen. Psychiatry41, 7280 (1984).

11. Owens, N. & McGorry, P. D. Seasonality of symptom onset in rst-episodeschizophrenia. Psychol. Med. 33, 163167 (2003).

12. Watson, C. G., Kucala, T., Tilleskjor, C. & Jacobs, L. Schizophrenic birthseasonality in relation to the incidence of infectious diseases and temperatureextremes. Arch. Gen. Psychiatry 41, 8590 (1984).

13. Fisman, D. N. Seasonality of infectious diseases. Annu. Rev. Public Health 28,127143 (2007).

14. Dowell, S. F. Seasonal variation in host susceptibility and cycles of certaininfectious diseases. Emerg. Infect. Dis. 7, 369374 (2001).

15. Kasahara, A. K., Singh, R. J. & Noymer, A. Vitamin D (25OHD) serumseasonality in the United States. PloS ONE 8, e65785 (2013).

16. Bunger, M. K. et al. Mop3 is an essential component of the master circadianpacemaker in mammals. Cell 103, 10091017 (2000).

17. Kondratov, R. V., Kondratova, A. A., Gorbacheva, V. Y., Vykhovanets, O. V. &Antoch, M. P. Early aging and age-related pathologies in mice decient inBMAL1, the core componentof the circadian clock. Genes Dev. 20, 18681873(2006).




18. Harmer, S. L. et al. Orchestrated transcription of key pathways in Arabidopsisby the circadian clock. Science 290, 21102113 (2000).

19. Meijer, J. H., Michel, S., Vanderleest, H. T. & Rohling, J. H. Daily and seasonaladaptation of the circadian clock requires plasticity of the SCN neuronalnetwork. Eur. J. Neurosci. 32, 21432151 (2010).

20. Dibner, C., Schibler, U. & Albrecht, U. The mammalian circadian timingsystem: organization and coordination of central and peripheral clocks. Annu.Rev. Physiol. 72, 517549 (2010).

21. Azzi, A. et al. Circadian behavior is light-reprogrammed by plastic DNAmethylation. Nat. Neurosci. 17, 377382 (2014).

22. Dyar, K. A. et al. Muscle insulin sensitivity and glucose metabolism arecontrolled by the intrinsic muscle clock. Mol. Metabol. 3, 2941 (2014).

23. Zhang, R., Lahens, N. F., Ballance, H. I., Hughes, M. E. & Hogenesch, J. B.A circadian gene expression atlas in mammals: Implications for biology andmedicine. Proc. Natl Acad. Sci.USA 111, 1621916224 (2014).

24. Shackelford, P. G. & Feigin, R. D. Periodicity of susceptibility to pneumococcalinfection: inuence of light and adrenocortical secretions. Science 182, 285287(1973).

25. Curtis, A. M., Bellet, M. M., Sassone-Corsi, P. & ONeill, L. A. Circadian clockproteins and immunity. Immunity 40, 178186 (2014).

26. Silver, A. C., Arjona, A., Walker, W. E. & Fikrig, E. The circadian clock controlstoll-like receptor 9-mediated innate and adaptive immunity. Immunity 36,251261 (2012).

27. Storch, K. F. et al. Extensive and divergent circadian gene expression in liverand heart. Nature 417, 7883 (2002).

28. Nguyen, K. D. et al. Circadian gene Bmal1 regulates diurnal oscillations ofLy6C(hi) inammatory monocytes. Science 341, 14831488 (2013).

29. Beyerlein, A., Wehweck, F., Ziegler, A. G. & Pueger, M. Respiratory infectionsin early life and the development of islet autoimmunity in children at increasedtype 1 diabetes risk: evidence from the BABYDIET Study. JAMA Pediatr. 167,800807 (2013).

30. Unsal-Kacmaz, K., Mullen, T. E., Kaufmann, W. K. & Sancar, A. Coupling ofhuman circadian and cell cycles by the timeless protein. Mol.Cell Biol. 25,31093116 (2005).

31. Yu, X. et al. TH17 cell differentiation is regulated by the circadian clock. Science342, 727730 (2013).

32. Barnes, P. J. Anti-inammatory actions of glucocorticoids: molecularmechanisms. Clin. Sci. 94, 557572 (1998).

33. Lamia, K. A. et al. Cryptochromes mediate rhythmic repression of theglucocorticoid receptor. Nature 480, 552556 (2011).

34. Bjornsdottir, U. S. et al. Pathways activated during human asthma exacerbationas revealed by gene expression patterns in blood. PloS ONE 6, e21902 (2011).

35. Ferreira, R. C. et al. A type I interferon transcriptional signature precedesautoimmunity in children genetically at-risk of type 1 diabetes. Diabetes 63,25382550 (2014).

36. Bronson, F. H. Are humans seasonally photoperiodic? J. Biol. Rhythms 19,180192 (2004).

37. Creekmore, A. L., Ziegler, Y. S., Boney, J. L. & Nardulli, A. M. Estrogen receptoralpha regulates expression of the breast cancer 1 associated ring domain 1(BARD1) gene through intronic DNA sequence. Mol.Cell Endocrinol. 267,106115 (2007).

38. Abbas, A. R., Wolslegel, K., Seshasayee, D., Modrusan, Z. & Clark, H. F.Deconvolution of blood microarray data identies cellular activation patterns insystemic lupus erythematosus. PLoS ONE 4, e6098 (2009).

39. De Jong, S. et al. Seasonal changes in gene expression represent cell-typecomposition in whole blood. Hum. Mol. Genet. 23, 27212728 (2014).

40. Moore, S. E. et al. Effect of month of vaccine administration on antibodyresponses in The Gambia and Pakistan. Trop. Med. Int. Health 11, 15291541(2006).

41. Kasasa, S. et al. Spatio-temporal malaria transmission patterns in Navrongodemographic surveillance site, northern Ghana. Malaria J. 12, 63 (2013).

42. Schenker, M. et al. Early expression and high prevalence of islet autoantibodiesfor DR3/4 heterozygous and DR4/4 homozygous offspring of parents withType I diabetes: the German BABYDIAB study. Diabetologia 42, 671677(1999).

43. Hurst, S. M. et al. Il-6 and its soluble receptor orchestrate a temporal switch inthe pattern of leukocyte recruitment seen during acute inammation. Immunity14, 705714 (2001).

44. Jones, S. A., Horiuchi, S., Topley, N., Yamamoto, N. & Fuller, G. M. The solubleinterleukin 6 receptor: mechanisms of production and implications in disease.FASEB J. 15, 4358 (2001).

45. Ferreira, R. C. et al. Functional IL6R 358Ala allele impairs classical IL-6receptor signaling and inuences risk of diverse inammatory diseases. PLoSGenet. 9, e1003444 (2013).

46. Emery, P. et al. IL-6 receptor inhibition with tocilizumab improves treatmentoutcomes in patients with rheumatoid arthritis refractory to anti-tumournecrosis factor biologicals: results from a 24-week multicentre randomisedplacebo-controlled trial. Ann. Rheum. Dis. 67, 15161523 (2008).

47. Cutolo, M. Rheumatoid arthritis: circadian and circannual rhythms in RA. Nat.Rev. Rheumatol. 7, 500502 (2011).

48. Knip, M. et al. Environmental triggers and determinants of type 1 diabetes.Diabetes 54(Suppl 2): S125S136 (2005).

49. Thompson, D., Pepys, M. B. & Wood, S. P. The physiological structure ofhuman C-reactive protein and its complex with phosphocholine. Structure 7,169177 (1999).

50. Querec, T. D. et al. Systems biology approach predicts immunogenicity of theyellow fever vaccine in humans. Nat. Immunol. 10, 116125 (2009).

51. Li, S. et al. Molecular signatures of antibody responses derived from asystems biology study of ve human vaccines. Nat. Immunol. 15, 195204(2014).

52. Nakaya, H. I. et al. Systems biology of vaccination for seasonal inuenza inhumans. Nat. Immunol. 12, 786795 (2011).

53. Grundberg, E. et al. Mapping cis- and trans-regulatory effects across multipletissues in twins. Nat. Genet. 44, 10841089 (2012).

54. Mohamed-Ali, V. et al. Subcutaneous adipose tissue releases interleukin-6,but not tumor necrosis factor-alpha, in vivo. J. Clin. Endocrinol. Metab. 82,41964200 (1997).

55. Roederer, M. et al. The genetic architecture of the human immune system:a bioresource for autoimmunity and disease pathogenesis. Cell 161, 387403(2015).

56. Orru, V. et al. Genetic variants regulating immune cell levels in health anddisease. Cell 155, 242256 (2013).

57. Wium-Andersen, M., rsted, D., Nielsen, S. & Nordestgaard, B. ELevatedc-reactive protein levels, psychological distress, and depression in 73 131individuals. JAMA Psychiatry 70, 176184 (2013).

58. Khandaker, G. M., Pearson, R. M., Zammit, S., Lewis, G. & Jones, P. B.Association of serum interleukin 6 and c-reactive protein in childhood withdepression and psychosis in young adult life: A population-based longitudinalstudy. JAMA Psychiatry 71, 11211128 (2014).

59. Brandes, M., Klauschen, F., Kuchen, S. & Germain, R. N. A systems analysisidenties a feedforward inammatory circuit leading to lethal inuenzainfection. Cell 154, 197212 (2013).

60. Yamaguchi, Y. et al. Mice genetically decient in vasopressin V1a and V1breceptors are resistant to jet lag. Science 342, 8590 (2013).

61. Bach, J. F. The effect of infections on susceptibility to autoimmune and allergicdiseases. N. Engl. J. Med. 347, 911920 (2002).

62. Archer, S. N. et al. Mistimed sleep disrupts circadian regulation of the humantranscriptome. Proc. Natl Acad. Sci. USA 111, E682E691 (2014).

63. Tamarkin, L., Baird, C. J. & Almeida, O. F. Melatonin: a coordinating signal formammalian reproduction? Science 227, 714720 (1985).

64. Lu, W., Meng, Q. J., Tyler, N. J., Stokkan, K. A. & Loudon, A. S.A circadian clock is not required in an arctic mammal. Curr. Biol. 20,533537 (2010).

65. Saenz de Miera, C. et al. A circannual clock drives expression of genes centralfor seasonal reproduction. Curr. Biol. 24, 15001506 (2014).

66. Tanikawa, C. et al. Genome wide association study of age at menarche in theJapanese population. PloS ONE 8, e63821 (2013).

67. Perry, J. R. et al. Parent-of-origin-specic allelic associations among 106genomic loci for age at menarche. Nature 514, 9297 (2014).

68. Stettler, C. et al. Serum amyloid A, C-reactive protein, and retinalmicrovascular changes in hypertensive diabetic and nondiabetic individuals: anAnglo-Scandinavian Cardiac Outcomes Trial (ASCOT) substudy. Diabetes Care32, 10981100 (2009).

69. Langfelder, P. & Horvath, S. WGCNA: an R package for weighted correlationnetwork analysis. BMC Bioinformatics 9, 559 (2008).

70. Wang, J., Duncan, D., Shi, Z. & Zhang, B. WEB-based GEne SeTAnaLysis Toolkit (WebGestalt): update 2013. Nucleic Acids Res. 41,W77W83 (2013).

AcknowledgementsWe gratefully acknowledge the participation of all BABYDIET families, patients andcontrol subjects. We thank Dr Sandra Hummel and Dr Maren Puger for coordinationof the BABYDIET study, D. Christiane Winkler for data management, S. Krause,M. Schulz, C. Matzke, A. Wosch and A. Gavrisan for sample processing, preparation ofPBMC samples, C. Peplow for consent and regulatory issues, and all paediatricians andfamily doctors in Germany for participating in the BABYDIET Study. We gratefullyacknowledge the participation of all NIHR Cambridge BioResource volunteers. We thankNeil Walker, and the Cambridge BioResource staff for their help with volunteerrecruitment. We thank members of the Cambridge BioResource SAB and ManagementCommittee for their support of our study and the National Institute for Health ResearchCambridge Biomedical Research Centre for funding. Access to Cambridge BioResourcevolunteers and their data and samples is governed by the Cambridge BioResource SAB.Documents describing access arrangements and contact details are available at http://www.cambridgebioresource.org.uk/. We would also like to thank Professor PatriciaMunroe and The Anglo-Scandinavian Cardiac Outcomes Trial (ASCOT) investigatorsfor their C-reactive protein data and advice. We also thank the population of West Kiang,




The Gambia for their participation, as well as our laboratory technicians, eld assistants,nurses, data entry clerks and other staff at MRC Keneba, The Gambia, involved in theKeneba Biobank. Many thanks also to Kerry Jones, Sophie E Moore, Bai Lamin Dondeh,and Sophie Hawkesworth for their input in setting up the Keneba Biobank. The Gambianstudy providing data for analysis was supported by core funding MC-A760-5QX00 to theInternational Nutrition Group by the UK Medical Research Council (MRC) and the UKDepartment for the International Development (DFID) under the MRC/DFID Con-cordat agreement. This work was supported by the JDRF UK Centre for Diabetes-Genes,Autoimmunity and Prevention (D-GAP; 4-2007-1003), the JDRF (9-2011-253), theWellcome Trust (WT061858/091157), the National Institute for Health ResearchCambridge Biomedical Research Centre (CBRC) and the Medical Research Council(MRC) Cusrow Wadia Fund. The research leading to these results has received fundingfrom the European Unions 7th Framework Programme (FP7/20072013) under grantagreement no.241447 (NAIMIT). The Cambridge Institute for Medical Research (CIMR)is in receipt of a Wellcome Trust Strategic Award (WT100140). X.C.D. was a Universityof Cambridge/Wellcome Trust Infection and Immunity PhD student. R.C.F. is funded bya JDRF post-doctoral fellowship (3-2011-374). C.W. and H.G are funded by theWellcome Trust (WT089989). The BABYDIET study was supported by grants from theDeutsche Forschungsgemeinschaft (DFG ZI-310/14-1 to-4), the JDRF (JDRF 17-2012-16and 1-2006-665) and the German Center for Diabetes Research (DZD e.V.). E.B. issupported by the DFG Research Center and Cluster of ExcellenceCenter forRegenerative Therapies Dresden (FZ 111). We also thank Joanna Rorbach, TomasCastro Dopico, Nikolas Pontikos, Charles Bell and Frank Waldron-Lynch for helpfuldiscussions.

Author contributionsX.C.D. had the idea for the study. X.C.D., M.E., C.W. and J.A.T. designed experimentsand analysed the data. M.E., H.G., N.C. and O.S.B. performed statistical analyses. C.W.supervised the statistical analyses. X.C.D. wrote the manuscript with contributions from

M.E., E.B., A.-G.Z., C.W. and J.A.T.; R.C.F., M.L.P. and D.J.S. performed wet labor in silico experiments and assisted with data processing. A.J.F., B.J.H. and A.M.P.were involved in the generation and analysis of blood count data from The Gambia.A.-G.Z. is the PI of the BABYDIET study and takes responsibility of the integrityof the data; A.-G.Z. and E.B. designed the BABYDIET protocol, provided samplesand clinical data.

Additional informationAccession codes: Mircoarray data have been deposited in ArrayExpress under accessioncode E-MTAB-1724.

Supplementary Information accompanies this paper at http://www.nature.com/naturecommunications

Competing nancial interests: The authors declare no competing nancial interests.

Reprints and permission information is available online at http://npg.nature.com/reprintsandpermissions/

How to cite this article: Dopico, X. C. et al. Widespread seasonal gene expressionreveals annual differences in human immunity and physiology. Nat. Commun. 6:7000doi: 10.1038/ncomms8000 (2015).

This work is licensed under a Creative Commons Attribution 4.0International License. The images or other third party material in this

article are included in the articles Creative Commons license, unless indicated otherwisein the credit line; if the material is not included under the Creative Commons license,users will need to obtain permission from the license holder to reproduce the material.To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/




title_linkResultsSeasonal ARNTL expression in the immune systemWidespread seasonal gene expression in the immune systemCommon seasonal genes in the immune system

Figure1Seasonal mRNA expression in the peripheral human immune system.Relative expression profiles of seasonal genes (fitted values of the cosinor model). (a) ARNTL expression was increased in the summer months of June, July and August (ANOVA, 2^2 , P=1Seasonal cellular remodelling of the human immune system

Figure2Widespread seasonal mRNA expression in the human immune system.Relative expression profiles of seasonal genes (fitted values of the cosinor model). (a) A total of 5,136 genes (sim23percnt of the protein-coding genome) were identified as having seaFigure3Seasonal gene expression in geographically distinct cohorts.(a) Seasonality was also observed in PBMCs collected from T1D patients in the United Kingdom (n=236 individuals). A total of 1,697 genes were seasonal in this data set. (b) The previouslyFigure4Seasonal changes in the cellular composition of human peripheral blood.(a) Expression levels of 13 genes that have been used to identify different blood cell types among total PBMCs were strongly correlated (positively and negatively) with seasonaSeasonal differences in human immunitySeasonal gene expression in subcutaneous adipose tissue

DiscussionFigure5Inflammatory responses predominate the immune system in Europe.(a) Co-regulated seasonal gene modules were generated to analyse differences in immune function by season: eight winter modules and three summer modules were generated. (b) Two modulesMethodsStudy subjects and human samples

Figure6Seasonal gene expression in subcutaneous adipose tissue.In a collection of 856 female adult donors from the United Kingdom, 4,027 genes were found to be seasonal in adipose tissue. As observed in PBMCs, two distinct anti-phasic profiles were preseGene expression analysis in BABYDIET and T1D PBMCsClimatic data for modelling seasonal gene expressionSelf-reported infections in BABYDIET cohortSoluble IL-6 receptor ELISACambridge BioResource full blood count data (UK cohort)Full blood count data from The GambiaC-reactive proteinStatistical analysis of the data setsDefinition of winter and summer seasonal genes in BABYDIETNetwork and functional analysis of the seasonal genes identified in BABYDIETAnalysis of the self-reported infections data of BABYDIET cohortIdentification of common seasonal genesSeasonal analysis of The Gambian full blood count dataNote added in proof

KohlerM.Marin-MoratallaN.JordanaX.AanesR.Seasonal bone growth and physiology in endotherms shed light on dinosaur physiologyNature4873583612012YanovskyM. J.KayS. A.Molecular basis of seasonal time measurement in ArabidopsisNature4193083122002EblingF. J.OnWe gratefully acknowledge the participation of all BABYDIET families, patients and control subjects. We thank Dr Sandra Hummel and Dr Maren Pflger for coordination of the BABYDIET study, D. Christiane Winkler for data management, S. Krause, M. Schulz, CACKNOWLEDGEMENTSAuthor contributionsAdditional information